Languages

Approximating the Speed of an Object and its Distance using OpenCV in Python

In this tutorial, we will learn how to use OpenCV to detect the frontal face of an object, find the object's distance in relation to the camera, then calculate the speed.

Note that we will use the laptop webcam as our tool.

Prerequisites
Referral face object image
Distance and speed approximation using the frontal face object
Getting started
Installations of external libraries
Detection of the frontal face
Determining the distance
Determining the speed
Conclusion

Prerequisites

You need to be conversant with Python as a programming language. To get started with Python basics, refer to this tutorial link below on Python.
You need to have Pycharm pre-installed since it will be our IDE working environment.

Referral face object image

This image shows how a frame of the frontal face is detected, and we will use it as our reference image later in the program.

ref-image

We will also use the haarcascade_frontalface_default.xml module to detect our face object. To get this module from Github, use this link.

Distance and speed approximation using the frontal face object

Getting started

Start your Pycharm IDE to create a new project. As shown in the image below.

startup

We will name our project DistanceVelocity on the open menu. We then select our base interpreter to python3.10 latest. You can use any version of the Python base interpreter as well. After making the suitable selections, we click on the create button to launch our project.

Installations of external libraries

The external library we will use in this section is OpenCV. In your working environment, there are different buttons in this window. Click on the terminal button to open the terminal interface. We will download and install our OpenCV from here. Copy the command below and paste it to the terminal for a successful installation.

  pip install OpenCV-Python

OpenCV is a powerful tool for aiding computer vision functions and related problems. It is also used in the processing of images and real-time videos.

Detection of the frontal face

We will have to use the haarcascade_frontalface_default.xml to detect the frontal face. Then, save it in the exact location of the main program. Next, we will work with our webcam, which by default is 0 during the calling function. Of course, 1 can also be used when dealing with an external camera, but we will use the default value for our case.

Then we will create a function for returning the detected face object coordinates of the rectangular frame. We then convert the RGB image into a gray-scale. It requires an image parameter for scaling up or down the image for better output.

High CPU processing power is required for this to be achieved. So we prefer to use the standard values. Let's dive into the coding section for a better understanding.

import cv2  
  
# This is the distance from camera to face object  
DECLARED_LEN = 30 # cm  
# width of the object face  
DECLARED_WID = 14.3 # cm  
# Definition of the RGB Colors format 
GREEN = (0, 255, 0)  
RED = (255, 0, 0)  
WHITE = (255, 255, 255)
#Defining the fonts family, size, type  
fonts = cv2.FONT_HERSHEY_COMPLEX  
# calling the haarcascade_frontalface_default.xml module for face detection.  
face_detector = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")  
  
  
def face_data(image):  
  
    face_width = 0  
  gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)  
    #We use 1.3 for less powerful processors but can increase it according to your processing power of your machine.
    faces = face_detector.detectMultiScale(gray_image, 1.3, 5) 
    #getting the rectangular frame 
    for (x, y, h, w) in faces:  
        cv2.rectangle(image, (x, y), (x + w, y + h), WHITE, 1)  
        face_width = w  
  
    return face_width  
#We use 0 in the VideoCapture function since that calls the default camera, the webcam.  
cap = cv2.VideoCapture(0)  
while True:  
    _, frame = cap.read()  
    face_width_in_frame = face_data(frame)  
    cv2.imshow("frame", frame)  
    #The string 'q' is will be used for stopping and quiting  
  if cv2.waitKey(1)==ord("q"):  
        break  
cap.release()  
cv2.destroyAllWindows()

Below is the image of the expected output. frontalFace

Determining the distance

To get the distance, we will have to create more functions. The first function to be created is the focal length finder.

Case 1: Finding the focal length

The defined function will calculate the focal length thus getting the distance between the lens and the CMOS sensor.

1st parameter to be used is Determined_Distance(int): It is the distance measured from the object to the Camera while Capturing a Reference image.
2nd parameter to be used is Actual_Width(int): This is the real width of the object, in the real world, for instance, my face width is = 14.3 centimeters).
3rd parameter to be used is Width_In_Image(int): This is the object width in the frame/image in our case in the reference image (found by Face detector)
Returning as decimal values.

Below is the function code.

def focal_length(determined_distance, actual_width, width_in_rf_image):  
    focal_length_value = (width_in_rf_image * determined_distance) / actual_width  
    return focal_length_value

Case 2: Finding the distance

This is the second function to be created, the distance finder function. This function approximates the distance between the face object and camera using defined arguments.

1st parameter to be used is focal_length(float): Returned by the focal_length_Finder function.
2nd parameter to be used is Actual_Width(int): It is the real width of the object. For example, my face is approximately 14cm in the real world.
3rd parameter to be used is object_Width_Frame(int): The broadness of object in the image (frame in our case, returning Video visual feed). Returning Distance as decimal values Distance(float): distance Estimated.

def distance_finder(focal_length, real_face_width, face_width_in_frame):  
    distance = (real_face_width * focal_length) / face_width_in_frame  
    return distance

Case 3: Reading reference images from the directory

We will open our reference image then store it in a variable. Further illustrations are implemented in the code below.

ref_image = cv2.imread("Ref_image.png")  
  
ref_image_face_width = face_data(ref_image) 
#Processing our called reference image
focal_length_found = focal_length(DECLARED_LEN, DECLARED_WID, ref_image_face_width)  
print(focal_length_found)  
cv2.imshow("ref_image", ref_image)  
  
while True:  
    _, frame = cap.read()  
  
    # calling face_data function  
  face_width_in_frame = face_data(frame)  
    # finding the distance by calling function Distance  
  if face_width_in_frame != 0:  
        Distance = distance_finder(focal_length_found, DECLARED_WID, face_width_in_frame)  
        # Writing Text on the displaying screen  
  cv2.putText(  
            frame, f"Distance = {round(Distance,2)} CM", (50, 50), fonts, 1, (WHITE), 2  
  )  
    cv2.imshow('frame', frame)  
    if cv2.waitKey(1) == ord("q"):  
        break  
cap.release()  
cv2.destroyAllWindows()

Your output should be as shown in the image below.

distance

Determining the speed

To find the speed, we will first have to import the time module since speed is the distance in relation to the time taken. Then we will initialize time-related variables.

Below is the code snippet as an example.

import time  
# declaration of variables  
initialTime = 0  
initialDistance = 0  
changeInTime = 0  
changeInDistance = 0  
  
listDistance = []  
listSpeed = []

Case 1: Finding the speed

This function takes the covered distance and time taken as parameters and returns the speed.

def speedFinder(covered distance, timeTaken):  
  
    speed = coveredDistance / timeTaken  
  
    return speed

Case 2: Finding average speed

We will start by finding the length of the list. Followed by calculating the number of items to find its average. Get the list of most recent items of the list to find the average of the selected items in the list. Then we return the average.

def averageFinder(completeList, averageOfItems):  
    lengthOfList = len(completeList)  
    selectedItems = lengthOfList - averageOfItems  
    selectedItemsList = completeList[selectedItems:]  
    average = sum(selectedItemsList) / len(selectedItemsList)  
  
    return average

Case 3: Calling of the pre-defined functions

Below is the code demonstrating the function calls.

while True:  
    _, frame = cap.read()  
  
    # calling face_data function  
  face_width_in_frame = face_data(frame)  
    # finding the distance by calling function Distance  
  if face_width_in_frame != 0:  
        Distance = Distance_finder(  
            Focal_length_found, DECLARED_WID, face_width_in_frame)  
        listDistance.append(Distance)  
        averageDistance = averageFinder(listDistance, 2)  
  
        # converting centimeters into meters  
  distanceInMeters = averageDistance/100  
  
  if initialDistance != 0:  
            # getting the  difference of the distances  
  changeInDistance = initialDistance - distanceInMeters  
  changeInTime = time.time() - initialTime  
  
            # calculating the speed  
  speed = speedFinder(  
                coveredDistance=changeInDistance, timeTaken=changeInTime)  
            listSpeed.append(speed)  
            averageSpeed = averageFinder(listSpeed, 10)  
            if averageSpeed < 0:  
                averageSpeed = averageSpeed * -1  
  # filling the progressive line dependent on the speed.  
  speedFill = int(45+(averageSpeed) * 130)  
            if speedFill > 235:  
                speedFill = 235  
  cv2.line(frame, (45, 70), (235, 70), (0, 255, 0), 35)  
            # speed dependent line  
  cv2.line(frame, (45, 70), (speedFill, 70), (255, 255, 0), 32)  
            cv2.line(frame, (45, 70), (235, 70), (0, 0, 0), 22)  
            # print()  
  cv2.putText(  
                frame, f"Speed: {round(averageSpeed, 2)} m/s", (50, 75), fonts, 0.6, (0, 255, 220), 2)  
  
            # print(speed)    
        initialTime = time.time()  
  
    # Writing Text on the displaying screen  
  cv2.line(frame, (45, 25), (255, 25), (255, 0, 255), 30)  
        cv2.line(frame, (45, 25), (255, 25), (0, 0, 0), 22)  
        cv2.putText(  
            frame, f"Distance = {round(distanceInMeters,2)} m", (50, 30), fonts, 0.6, WHITE, 2)  
    # recording the video  
  Recorder.write(frame)  
    cv2.imshow("frame", frame)  
    if cv2.waitKey(1) == ord("q"):  
        break

Below is the output.

output